Strange Vray CPU Rendertimes - Simple Scene takes ages and i don't know whats causing it

[olika]

Hi Vray-Team,

i just came across something which seems strange to me.
While i was working on an animation i realised that thing are kinda slow.
So i started investigating, double checked the scene and whatnot. The thing is that the renderspeed seems too slow in the latest vray for my experience and compared to the past.
I stripped it down to a very simple scene. Just a CAD PArt (not that heavy) a simple domelight and a simple metal shader (edgetex is the only fancy on that shader).
The results are somehow worrying me as vray needs more than 6 minutes for that simple hd frame?
Lighting seems ok, when i render it with a simple grey it rushes through in 30 seconds....

Reaching out for your support here, because i already checked pretty everything i can imagine

- Using different, smaller hdris
- Trying Dome vs. Env, Light slots
- Reduce global reflections to 3
- Reinstall Vray
- Simplify shader
- turn off refl. caustics
- Merge in a new scene
- etc.

Looking for something i miss right now, or just one checkbox i`m not thinking about, or is something else going wrong here?
I'm not sure but rendertime over 6 minutes for a render like this seems ridiciolous to me, am i wrong?

System specs are:
- Latest Vray
- Win 10
- Intel i9-7920x 2,9Ghz
- 128 gb ram

I also attached the renders and scene, hoping somebody can check, and hopefully loosen my break on vray right now!
Thx in advance, i hope i gave all information needed!

[kosso_olli]

Your slowdown is caused by edge texture, most definately. Assign it to the grey override as well and the difference should be much smaller.

[olika]

Hi kosso_olli
thx for helping!
Totally true what you say and i'm also aware of that. I compared it, there's obviously a difference but i'm still think there's something else going on here.
Grey: from 37sec to 34 sec, Metal from 6min to 4min... hmmm is 4 minutes a good time?

Do you remember the days a few years ago when vray switched to the new sample method and there was a difference in sample numbers for the dome when you do i.e. interior shot or you do that simple "just a object and only a dome" thing?
Now that all is so simplified and luckily we don't have to deal with that samples anymore.
I still wondering if i should change settings when i do scenes like this? Just assuming maybe with standard settings it's optimised more to archiz right now?
I mean domelight is setup to do a good job lighting an interior and maybe for my scenes i.e. adaptive dome maybe not good or maybe try going the old school way and ressurect domelight samples?

[Muhammed_Hamed]

4 minutes is kind of okey for this on 12 cores I think, considering your blurry reflections on this piece
I don't recommend going to the old workflow of setting up local samples, it is not going to improve your render times. Vray defaults are for all cases, not just interiors
What render settings are you using for this?
adaptive or bucket mode?
Noise threshold? did you change any defaults like LC settings?
For animation you usually can go higher on noise threshold if you plan on using denoiser, you can get render time to 2 minutes or so
And I would recommend running Vray benchmark, just to make sure your CPU is working as expected.
You should get around 20k points in the CPU test.

[piotrus3333]

topology seems pretty messy (long thin triangles, this highlights are giveaway). this plus VRayEdgesTex might be causing slowdown.

[fixeighted]

I think that is the reflection of the circular piece on top. If it's a direct import maintaining the underlying brep objects then it shouldn't really suffer from triangulation issues.
Not guaranteed though and it was my first thought also, and that possibly converting it to a sensible poly could help, or not....depends.

Personally I would think this render time is fine considering the quality of the glossies/edgetex etc.

[piotrus3333]

I think that is the reflection of the circular piece on top. If it's a direct import maintaining the underlying brep objects then it shouldn't really suffer from triangulation issues.
Not guaranteed though and it was my first thought also, and that possibly converting it to a sensible poly could help, or not....depends.

Personally I would think this render time is fine considering the quality of the glossies/edgetex etc.

yep, you’re right - just proper reflection.
any chance to see the topo?

[fixeighted]

The scene file is provided in the link Just having a look now...

[olika]

Hi Muhammed_Hamed piotrus3333 and fixeighted thanks for joining in!
I also played around the whole morning with a good friend who also looked at it and tried to help.

We tried reimporting it, then used few different ways of welding stuff together, but in the end i think the model is quite fine in the sense how the mesh behaves. There was not a big difference in the different imports at all. Also not with the rendertimes.
So yeah the mesh could be way cleaner but it seems that doesnt make a difference. But hey thanks i already learned today that you could have your bodyobjects kept underneath and that may make some things easier in the future!

And jup if i don't really need to change settings different from standards i usually won't do it!

[fixeighted]

I have an i7-5960x 3ghz and I achieved an enormous increase in render speed by simply using a different hdr. All the other settings are the same I believe. Here's the render and the hdr to test https://www.dropbox.com/s/ivo70bc4gy...SMALL.hdr?dl=0

EDIT: I meant to say that my crappy old processor took 14.24 mins to do this. Before swapping out the hdr it was looking like taking at least 35mins...maybe longer....couldn't be arsed to let it complete

[^Lele^]

There's nothing wrong with the way this is rendered, that i can see.
If you look at the samplerate RE when rendering the metal, you'll notice sampling is concentrated on highlights which are either small, or strong, or both.
That's expected, and it will depend on the topology, material, AND light source (or HDRI): small and intense will make for longer sampling.
You can find more talk about this class of shading problems here and here

[piotrus3333]

heh, even welding all the verts did not help (down to 55k on the bigger part).
I guess do not waste time on trying to improve it, just use denoiser.

[olika]

Thx for all the response!
^Lele^ jup indeed the samplerate RE looked fine all the time, that was one the things which makes me wonder. Normally that means buckets running.
When you have a strong hdri and it's having a though time sampling the highlights you can see that normally in the samplerate RE as redish areas.

Maybe I was just off because the last job i did was a large aerial environment shot with lots of forestpack foliage and whatnot and i had rendertimes around 10 minutes mark for this.
So could be i was just bummed in front of that simple thing Glossy is still glossy i guess in terms of the time it needs getting it out clean!

[^Lele^]

Yeah, if you consider the screen area, one such flat and glossy piece is likely a lot harder to sample than foliage in the distance with tenuous highlights...
Mileage will vary depending on a few things, but the amount of screen coverage for a particular portion of the scene (your glossy shader, or any other expensive effect) is normally a decent enough metric to gauge why it's taking the time it takes.
Same goes for the SR RE: in your case, you're also -conveniently- clamping max AA as V-Ray would call for more subdivs still to ensure that highlight had a noise level of 0.01, as set.
Big screen areas covered in red give you a hint of that happening, and of why it's taking a longer time to complete.

[fixeighted]

Did you compare times with the hdr I used? Your times should have come down loads and the highlights are mostly the same/similar.
So even if you still used yours but brought down the extremes of the ceiling it would look practically identical in term of the reflection but would take much less time, no?